Selection and Agglomeration Impact on Firm Productivity: A Study of Taiwan's Manufacturing Sector SYED HASAN, ALLEN KLAIBER AND IAN SHELDON OHIO STATE UNIVERSITY NARSC ANNUAL MEETING 2013
Significance of TFP Estimation Importance of Technology vs Capital in economic growth Abramovitz (1956)-Measure of our ignorance Solow (1957) Melitz (2003)-Survival and scope of the firm in the marketheterogeneous firms (and heterogeneity is thought of in terms of productivity)
Regional Productivity Rosenthal and Strange (2003)-Firms in large cities have high productivity Ciccone and Hall (1996)-Positive relation between density of economic activity and firm s productivity Large City- Above median employment density Clusters/Science Park-Public-private partnership that fosters knowledge flows among firms and contributes to regional economic growth Market failure- Innovations
Research Motivation To compare the relative productivity distributions of firms located in clusters to those located in large cities and small cities To compare industry level productivity distribution within the cluster What drives the productivity-agglomeration or Selection Policy Implications- Do they deliver
Research Methodology TFP Estimation Methods OLS Fixed effects IV Sources of Bias Simultaneity Selectivity Production Function y it = β 0 + β k k it + β l l it + ε it y it = β 0 + β k k it + β l l it + v it + u it ω it = β 0 + v it E x it ω it = 0
Semi-Parametric Estimation Olley and Pakes(1996)- semi-parametric method, based on proxy variables- i) the identification of a proxy variable, which is function of TFP, and ii) the definition of the conditions under which this function can be inverted in order to express TFP as a function of the proxy variable itself
Why Taiwan? Taiwan is now the home of many of the world's largest makers of computers and associated hardware. Its firms produce more than 50% of all chips, nearly 70% of computer displays and more than 90% of all portable computers (Economist, 2010).
Results-TFP Estimates Data: Emerging Markets Information Services, firm level, income statement and balance sheet. Industry classification at 3-digit NAICS level. OLS IV OP β k 0.37*** 0.56*** 0.29** β l 0.56*** 0.21*** 0.47** Deflate sales Remove outliers Data limitations Materials Energy Firm prices Multi-Product Bias When products have bigger technological -give up some product lines. Moreover, multi-product plants are shown to exit markets where production technologies are farthest away from their primary products.
Regional TFPs POPULATION DENSITY TFP-COUNTY MARKET
Summary Stats-Log TFP Stats BM SP AM N 840 1427 2388 mean 4.10692 8.32283 11.7669 max 8.70842 12.1029 17.0863 min -2.4334 1.00501 4.60511 IQR 1.23055 1.3496 1.41828
Inter-Industry comparison-technology intensive levels COMPUTER AND ELECTRONICS CHEMICAL MANUFACTURING
Agglomeration and Selection Localization-Henderson et al(1995)- Regional employment share of the specific industry Urbanization- Herfindahl-Hirschman Index which is 2 computed as s jrt where s is the employment j share of two digit manufacturing industry j Competition-Population density; diseconomies of scale or local demand
Self Selection Baldwin and Okubo (2006) show that high productivity firms self-select into large markets Heckman two-step estimator for selection models (Heckman, 1976; 1979) Robustness Checks Dummy variable for location Instrument the endogenous variable where z it = α 0 + αc it + ε it z it z it = 1 S prt = β 0 + β a A rt + β c X rt + υ prt S prt is the p-th percentile of region r at time t, Art are industry specific agglomeration variables for region r at time t, error term. X rt are the region-time specific control variables and prt is the
Agglomeration OR Selection Following Syverson (2004) and Combes et al (2012) i. Mean (median)-to check for relative shift ii. IQR -for dispersion Agglomeration and Selection in Science Park IQR MED 10-TILE LOC URB Lab Den LOC URB Lab Den LOC URB Lab Den 1.89*** -3.47*** -.07** 0.48*** -0.35*** 0.31*** -.70*** 1.84***.04*** (.10) (0.16) (.02) (0.01) (.06) (.00) (.06) (.09) (.01) iii. 10th percentile - for truncation/cut-off
Probability Density Simulations-To be done Simulations to suppress noise in TFP estimation TFP Distribution of empirical data Combes(2012) and Melitz & Ottaviano (2008) 0.45 0.4 0.35 0.3 Probability Density Function empirical loglogistic tlocationscale logistic gamma Normal Quantile Plots 0.25 ML estimates 0.2 0.15 0.1 0.05 0 4 6 8 10 12 14 16 18 Value
Results Firms in large cities have the highest level of productivity Firms located in Science parks usually have intermediate productivity levels in between large and small cities Firms productivity in Science parks may depend on the technology intensity of the production process
Conclusion Policy Implications Creating clusters may enhance productivity of technology intensive industry Clusters may turn out to be protective shields against competition in some cases
Comments:hasan.47@osu.edu